A regional progressive mesh provides support for real-time rendering of large-scale surfaces with locally adapting surface geometric complexity according to changing view parameters. The regional progressive mesh is constructed by subdividing an initial detailed mesh one or more times into multiple sub-regions as an iterative or recursive process. Each sub-region is separately simplified, and the localized transformations recorded in separate segments in a sequence of mesh refinement transformations that form the progressive mesh representation. The resulting regionalized organization of mesh refinement transformations reduces the working set of memory pages containing progressive mesh data needed for real-time view-dependent adaptation and rendering of the mesh surface. An exact approximate error measurement of a vertex split transformation also is defined as the maximum height deviation at enumerated vertices in the open neighborhood of the transformation relative to a regular triangulation of grid points, where the enumerated vertices include the grid points internal to the faces adjacent the split vertex and the grid line crossings internal to edges adjacent the split vertex.
|
14. A method comprising:
identifying a plurality of operations that, when applied to a simplified mesh, generate a more refined version of the simplified mesh; and
organizing representations of the operations into groups of plural operations according to within which region out of a plurality of regions of the simplified mesh the operations refine the mesh;
wherein at least two of the plurality of regions have a plurality of operations associated therewith.
20. A method of generating a mesh representation having a hierarchical region-based organization based on a mesh, the method comprising:
repeatedly subdividing the mesh into plural sub-regions until a stop state is met for at least a plurality of the sub-regions;
simplifying the plurality of the sub-regions of the mesh;
combining a plurality of the simplified sub-regions into a composite simplified mesh; and
storing the composite simplified mesh in the mesh representation wherein a plurality of mesh operations operate within at least one of the subregions.
18. A method of generating a progressive mesh representation having a hierarchical region-based organization based on a mesh, the method comprising:
for a plurality of sub-regions of the mesh at a level of detail:
identifying a plurality of operations for regenerating the sub-regions from simplified versions of the sub-regions; and
storing, as a group by sub-region, representations indicating the plurality of operations for regenerating the sub-regions from the simplified versions in the representation at a hierarchical level related to the level of detail for the sub-regions.
1. A computer-implemented method for generating a representation having a hierarchical region-based organization for a mesh, such method receiving an input mesh, the method comprising:
sub-dividing the input mesh into plural sub-regions;
identifying a plurality of operations per sub-region simplifying the sub-regions; and
storing, in the representation having a hierarchical region-based organization, data indicating a plurality of refinement operations derived from the plurality of operations per sub-region simplifying the sub-regions, wherein the storing comprises grouping the data indicating the plurality of refinement operations according to the sub-region within the input mesh whereat the refinement operations operate.
22. A method of producing a simplified version of a mesh, the method comprising, based on the overall detailed mesh as an initial input:
if a stop state has not been met for an input mesh:
subdividing the input meshing a plurality of sub-regions;
processing the sub-regions as input meshes, the processing comprising simplifying the input sub-regions;
combining the simplified sub-regions into a composite simplified mesh; and
processing the composite simplified mesh as an input mesh, the processing comprising simplifying the composite simplified mesh;
whereby the simplified version of the overall detailed mesh is generated without allocating storage based on the size of the overall detailed mesh and wherein a plurality of mesh operations operate within at least one of the subregions.
12. A computer-implemented method for generating a representation for a recursion input mesh, such method receiving a fully detailed mesh as an initial input, in which the recursion input mesh is stored in a memory having a size, the method comprising:
testing for a recursion stop state, such state being true or false, and wherein the stop state is determined at least in part by the size of the memory required to store the recursion input mesh;
if the state is true, then simplifying an input mesh of a current recursion of the method and falling back in the recursion; and
if the state is false, then sub-dividing the recursion input mesh into plural sub-regions, and recursing on each of the plural sub-regions wherein a plurality of mesh operations operate within at least one of the subregions.
23. A computer-readable medium comprising a data structure built via the following computer-implemented method for generating a representation having a hierarchical region-based organization for a mesh, such method receiving an input mesh, the method comprising:
sub-dividing the input mesh into plural sub-regions;
identifying a plurality of operations per sub-region simplifying the sub-regions; and
storing, in the representation having a hierarchical region-based organization, data indicating a plurality of refinement operations derived from the plurality of operations per sub-region simplifying the sub-regions, wherein the storing comprises grouping the data indicating the plurality of refinement operations according to the sub-region within the input mesh whereat the refinement operations operate.
2. The method of
3. The method of
4. The method of
combining the plural sub-regions into a single region; and
simplifying the single region.
5. The method of
6. The method of
testing for a stop state; and
repeatedly performing the sub-dividing until the stop state is met;
wherein the testing for the stop state includes detecting a mesh shape.
7. The method of
testing for a stop state; and
repeatedly performing the sub-dividing until the stop state is met;
wherein the testing for the stop state comprises detecting a size of at least one of the sub-regions, where size is determined at least in part by a number of vertices.
8. The method of
testing for a stop state; and
repeatedly performing the sub-dividing until the stop state is true;
wherein the stop state is true when at least one of the sub-regions has a level of detail below a predetermined threshold.
9. The method of
the representation represents a plurality of levels of detail; and
the data indicating the operations is further grouped according to a level of detail generated by the operations.
10. The method of
the operations are edge collapses; and
the data indicating the operations comprises information indicating vertex splits, wherein the vertex splits are inverses of the edge collapses.
11. A computer-readable medium comprising computer-executable instructions for performing the method of
13. The method of
15. A computer-readable medium comprising computer-executable instructions for performing the method of
16. The method of
17. The method of
generating a simplified base mesh;
storing the simplified base mesh in a data structure; and
storing the representations of the operations in the data structure according to a hierarchical region-based organization.
19. The method of
the operations are edge collapses; and
the representations indicating the operations represent vertex splits, wherein the vertex splits are inverses of the edge collapses.
|
A portion of this disclosure may contain copyrighted material. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights.
The present invention relates generally to geometric modeling using polygonal meshes for computer graphics, and more particularly relates to techniques for optimizing computer resource (e.g., memory, CPU, etc.) requirements for the piecewise manipulation of large-scale meshes.
Rendering real-time views on a complex model is a computation-intensive task. Current methods generally rely on encoding real-world objects in a complex three-dimensional geometric model which approximates the surfaces, textures and locations of the real-world objects. Objects are usually represented as a collection of polygons (but can be a mathematical description, fractal or the like) which are collected together to form a mesh having the shape and visual and/or tactile characteristics of the encoded real-world object. Realistic models of any reasonable size, e.g., simulating large-scale terrain meshes (see FIG. 2), can contain hundreds of meshes with millions of polygons to represent a realistic view of reality. Consequently, enormous computer resources are required to manipulate large meshes.
To simplify processing when rendering views of such detailed scenes, view-dependent progressive-meshes (VDPM) were developed. Hoppe shows a VDPM framework that represents an arbitrary delta-mesh (base mesh plus deltas required to produce a more detailed mesh) as a hierarchy of geometrically optimized transformations. A series of geomorph operations can be performed to convert between differing levels of detail of the progressive mesh. (For further information, see Hoppe, Progressive Meshes, Computer Graphics Proceedings, Annual Conference Series (ACM SIGGRAPH), pp. 99-108 (1996); Hoppe, View-Dependent Refinement of Progressive Meshes, ACM SIGGRAPH, pp. 189-198 (1997).)
For further information regarding techniques for constructing and using progressive meshes, view-dependent progressive meshes, and geomorphs, see: U.S. Pat. No. 5,963,209 for Encoding And Progressive Transmitting Of Progressive Meshes, bearing application Ser. No. 08/586,953 and filed Jan. 11, 1996; U.S. Pat. No. 5,966,133 for Selective Refinement Of Progressive Meshes, bearing application Ser. No. 08/797,501 and filed Feb. 7, 1997; U.S. Pat. No. 6,108,006 for View-Dependent Refinement Of Progressive Meshes, bearing application Ser. No. 08/826,570 and filed Apr. 3, 1997; U.S. Pat. No. 6,137,492 for Adaptive Refinement Of Progressive Meshes, bearing application Ser. No. 08/826,573 and filed Apr. 3, 1997; and U.S. Pat. No. 5,966,140 for a Method For Creating Progressive Simplicial Complexes, bearing application Ser. No. 08/880,090 and filed Jun. 20, 1997. These patents are incorporated herein by reference.
A problem with progressive mesh implementations is that they generally allocate storage based on the size of the original fully-detailed mesh. The Hoppe VDPM h framework alleviates some of these resource requirements by representing the original fully detailed mesh as a progressive mesh storing a simplified base mesh and a sequence of mesh refinement transformations (i.e., vertex splits). These refinements, when applied to the base mesh, exactly reconstruct the original fully detailed mesh.
To further reduce resource requirements, Hoppe defines a viewpoint with respect to the mesh, and implements a “selectively refined mesh” (SRM), which is a mesh that is altered based on satisfying certain viewing criteria. The viewpoint corresponds to what a user might see of a mesh, or to what is visible in a viewport. The result is view-dependent level of detail (LOD) applied to different portions of a mesh. For example, viewing conditions may depend on the location of a flight simulator pilot in a mesh model, so that mesh detail centers on those portions of the mesh near the pilot's plane. For a view-dependent LOD, vertices of the mesh are either coarsened or refined based on the view-dependent refinement criteria. Examples of view-dependent criteria affecting the occurrence of a vsplit (vertex split) operation is whether a vertex's neighborhood intersects the view frustum (see FIG. 7), has a Gauss map not strictly oriented away, or has a screen-projected deviation from M (the original fully detailed mesh) that exceeds a pre-determined pixel tolerance. The mesh alteration is either effected instantly, or geomorphed (interpolated) over time. Which method is used depends on whether an affected mesh region is entering, leaving, or staying within a particular viewing frustum (e.g. the view dependency). Only portions of a mesh remaining in view (i.e., within the frustum) need to be geomorphed to avoid “popping” the mesh alteration into place.
A significant problem with progressive mesh systems, however, is that they consume computer resources proportional to the size of the fully detailed mesh Mn. That is, in the prior art, if Mn has p faces, but has been simplified into a simpler mesh M0 having only q faces, where q<<p (much less than), memory and other computer resources are allocated on a scale proportional with the connectivity of the larger mesh Mn. For example, to compute a geomorph between two SRMs of the progressive mesh hierarchy, some prior art methods require computations involving all vertices, rather than on just those vertices needed for a view-dependent computation. Requiring resource allocation for inactive vertices is very inefficient.
To minimize resource requirements, the invention optimizes storage requirements by utilizing dynamic data structures for storing and manipulating a mesh that allocate storage based on the active vertices of the mesh. Thus, storage requirements fluctuate according to view-dependent criteria. Instead of statically storing mesh connectivity for an entire fully detailed mesh Mn (prior art), static data structures only encode the vertex hierarchy of the simplest mesh M0 and the refinement dependencies required to produce Mn. Separate dynamic data structures encode, according to changes in view-dependent parameters, vertices and morph states for an active mesh. This results in substantial resource savings when (as is typical) the fully detailed mesh is large, but the number of active vertices is just a small subset of the overall number of vertices.
However, if a mesh has a huge number of faces, such as for a large terrain mesh, even with dynamic data allocation, resource requirements may still be very substantial.
To overcome this problem, the original fully detailed mesh is subdivided into multiple arbitrarily shaped “start” regions, which when combined, represent the entire original mesh. For simplicity, it is assumed the regions are rectangular. When dividing the original mesh, each start region is recursively subdivided into smaller sub-regions, until a stop state is reached. Such a stop state may be based on region size, region complexity, resource requirements (e.g., memory or processing), or some other predetermined basis. When the recursion stop state is reached (assuming a simple recursive algorithm), each start region has been subdivided into many smaller mesh regions. Before falling back in the recursion process, each stop state region is simplified to a predetermined level of simplification. This corresponds to a standard progressive mesh computation of the most detailed (e.g. bottom) level of the mesh hierarchy.
The now-simplified stop state regions are stitched together to form a larger region. This undoes the region division of the previous recursive step. On falling back in the recursion this larger stitched region is then again simplified and stitched with other regions from this previous recursive level. Note that although the term “stitched” is used, the mesh may be stored such that only one copy of the mesh exists, so that simplifying the stop state regions directly simplifies the original mesh. So long as the simplification steps are recorded, the original mesh may be selectively replaced in the simplification process. Eventually all recursive levels are traversed, leaving a very simple mesh (M0) having only a few triangle faces.
During manipulation of the mesh hierarchy, selective refinement or coarsening operations can be constrained to operate only within a single sub-region. This is a natural restriction since a proper blocking, where no faces cross a region border, yields independent subregions. On multi-processor machines, or where operating systems programming environments provide multitasking or multi-threading services, mesh operations for each region can therefore be processed in parallel.
Additional features and advantages of the invention will be made apparent from the following detailed description of an illustrated embodiment which proceeds with reference to the accompanying drawings.
The file of this patent contains at least one drawing executed in color. The Patent and Trademark Office, upon request and payment of the necessary fee, will provide copies of this patent with color drawings.
The present invention is directed toward geometric modeling using polygonal meshes for computer graphics. In particular, a view-dependent progressive-mesh framework allows view-dependent level-of-detail control over arbitrary triangle meshes. Such view-dependent control over mesh rendering is particularly well-suited for interactive environments, such as 3D games presenting flights over background meshes, or fly-through reviews of multi-dimensional models (e.g., a real-time fly-through of a CAD/CAM project). In such situations, a viewer/user only sees a small portion of an entire mesh model, and within the visible portion of the mesh many regions are distant and can therefore be represented at a coarser level-of-detail.
The invention conserves computer resources by limiting most resource allocation to those areas of the mesh that are visible (or near) to the viewer/user. Known progressive-mesh techniques are extended to achieve temporal coherence through the runtime creation, in real-time, of geomorphs. Creating geomorphs at run-time allow the mesh to remain in a coarse state until such time as user movement requires refining portions of the mesh. The invention also conserves computer processing resources by distributing active-vertex evaluations and geomorph operations across multiple (possibly overlapping) operations.
For very large meshes, the invention can recursively subdivide these meshes into small mesh regions such that for each recursion level, as the recursion unwinds, the smaller regions are simplified and combined into larger regional meshes. After fall back to a previous recursion level, these larger meshes are then simplified and combined. This process repeats until all regions have been combined and simplified during recursion unwinding. The final result is a base mesh for a progressive mesh hierarchy, but where all refinement operations needed to obtain the fully detailed mesh have been determined from processing the arbitrarily sized regions. The refinement operations are grouped according to the regions defined for the original input mesh, thus allowing memory utilization to be optimized. An additional advantage of the invention is that the maximum error for a mesh region can be tracked as the region is simplified during preparation of the mesh hierarchy, thereby allowing refined rendering of some regions to be entirely avoided if its impact on an output device is determined to be insignificant (thus saving on having to allocate resources for such refinements).
For very large meshes exceeding available RAM, the initial partitioning of the mesh can be performed without loading the entire fully detailed mesh into memory. Instead, the mesh can be initially partitioned by performing a partial-load of mesh data into memory, or by manipulating mesh data in permanent storage (e.g. tape, disk or CD-RW). Region sizes can be selected so that an entire region will fit within a single page of memory, thus increasing memory access/efficiency.
FIG. 1 and the following discussion are intended to provide a brief, general description of a suitable computing environment in which the invention may be implemented. While the invention will be described in the general context of computer-executable instructions of a computer program that runs on a personal computer, those skilled in the art will recognize that the invention also may be implemented in combination with other program modules. Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Moreover, those skilled in the art will appreciate that the invention may be practiced with other computer system configurations, including hand-held devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, minicomputers, mainframe computers, and the like. The illustrated embodiment of the invention also is practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. But, some embodiments of the invention can be practiced on stand alone computers. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.
With reference to
The system bus may be any of several types of bus structure including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of conventional bus architectures (e.g., PCI, AGP, VESA, Microchannel, SSA, ISA, EISA, IEEE1394, Fibre Channel, and SCSI-FCP, etc.). The system memory includes read only memory (ROM) 24 and random access memory (RAM) 25. A basic input/output system (BIOS), containing the basic routines that help to transfer information between elements within the personal computer 20, such as during start-up, is stored in ROM 24.
The personal computer 20 further includes a hard disk drive 27, a magnetic disk drive 28, e.g., to read from or write to a removable disk 29, and an optical disk drive 30, e.g., for reading a CD-ROM disk 31 or to read from or write to other optical media. The hard disk drive 27, magnetic disk drive 28, and optical disk drive 30 are connected to the system bus 23 by a hard disk drive interface 32, a magnetic disk drive interface 33, and an optical drive interface 34, respectively. The drives and their associated computer-readable media provide nonvolatile storage of data, data structures, computer-executable instructions, etc. for the personal computer 20. Although the description of computer-readable media above refers to a hard disk, a removable magnetic disk and a CD, it should be appreciated by those skilled in the art that other types of media which are readable by a computer, such as magnetic cassettes, flash memory cards, digital video disks, Bernoulli cartridges, and the like, may also be used in the exemplary operating environment.
A number of program modules may be stored in the drives and RAM 25, including an operating system 35, one or more application programs 36, other program modules 37, and program data 38.
A user may enter commands and information into the personal computer 20 through a keyboard 40 and pointing device, such as a mouse 42. Other input devices (not shown) may include a microphone, joystick, game pad, satellite dish, scanner, or the like. These and other input devices are often connected to the processing unit 21 through a serial port interface 46 that is coupled to the system bus, but may be connected by other interfaces, such as a parallel port, game port or a universal serial bus (USB). A monitor 47 or other type of display device is also connected to the system bus 23 via an interface, such as a video adapter 48. In addition to the monitor, personal computers typically include other peripheral output devices (not shown), such as speakers and printers.
The personal computer 20 may operate in a networked environment using logical connections to one or more remote computers, such as a remote computer 49. The remote computer 49 may be a server, a router, a peer device or other common network node, and typically includes many or all of the elements described relative to the personal computer 20, although only a memory storage device 50 has been illustrated in FIG. 1. The logical connections depicted in
When used in a LAN networking environment, the personal computer 20 is connected to the local network 51 through a network interface or adapter 53. When used in a WAN networking environment, the personal computer 20 typically includes a modem 54 or other means for establishing communications over the wide area network 52, such as the Internet. The modem 54, which may be internal or external, is connected to the system bus 23 via the serial port interface 46. In a networked environment, program modules depicted relative to the personal computer 20, or portions thereof, may be stored -in the remote memory storage device. It will be appreciated that the network connections shown are exemplary and other means of establishing a communications link between the computers may be used.
In accordance with the practices of persons skilled in the art of computer programming, the present invention is described below with reference to acts and symbolic representations of operations that are performed by the personal computer 20, unless indicated otherwise. Such acts and operations are sometimes referred to as being computer-executed. It will be appreciated that the acts and symbolically represented operations include the manipulation by the processing unit 21 of electrical signals representing data bits which causes a resulting transformation or reduction of the electrical signal representation, and the maintenance of data bits at memory locations in the memory system (including the system memory 22, hard drive 27, floppy disks 29, and CD-ROM 31) to thereby reconfigure or otherwise alter the computer system's operation, as well as other processing of signals. The memory locations where data bits are maintained are physical locations that have particular electrical, magnetic, or optical properties corresponding to the data bits.
The problem with such a large mesh is that the general approach for rendering such surfaces is to exploit the traditional 3D graphics pipeline, which is optimized to transform and texture-map triangles. The graphics pipeline has two main stages: geometry processing and rasterization. Typically, the rasterization effort is relatively steady because the rendered surface has low depth complexity. In the worst case, the model covers the viewport, and the number of filled pixels is only slightly more than that in the frame buffer. Current graphics workstations (and soon, personal computers) have sufficient fill rate to texture-map the entire frame buffer at 30-72 Hz, even with advanced features like trilinear mip-map filtering and detail textures. Such 3D processing speed is largely due to the advent of inexpensive consumer-oriented 3D graphics boards optimized for processing 3D graphics languages used to encode the models. Consequently, geometry processing proves to be the bottleneck. (Note that the techniques discussed herein can be applied to speed-up hardware-based geometry processors, for 3D graphics languages such as OpenGL, which are designed to widen the processing bottleneck.)
Due to this bottleneck, in the prior art, complex meshes require specialized hardware in order to render them in real time, since most high-end graphics platforms can not process, in real time, the geometry for more than a small fraction of the 17 million triangles. The invention overcomes this problem by optimizing the representation of the complex mesh. A progressive mesh hierarchy, having multiple simpler meshes, is substituted in real-time for portions of the fully detailed (original/complex) mesh. Substitutions are made in real-time with respect with a viewer/user's changing viewing perspective (viewpoint) within the mesh, thus providing for real-time processing of complex meshes.
It is important to realize that when handling complex meshes, there is little point in rendering more triangles than there are pixels on an output device. For example, as noted, the
In order to simplify a fully detailed mesh Mn such as
The best representative meshes are potential meshes Mp having the least amount of error with respect to a more detailed mesh. Comparisons can be made between more-detailed intermediate meshes Mi+1, or more preferably, against the fully detailed mesh Mh. A heuristic h is employed to compute the error between a more detailed mesh and the potential mesh, and quantify the degree of similarity between the meshes. The potential mesh Mp having lowest error is selected to be the simplified mesh Mi for a given level i of the hierarchy. In other words, a potential mesh Mp results from applying collapse/split operations to a previous mesh, where the goal is to apply a heuristic to determine the “best” possible potential mesh. The term “heuristic” includes evaluating each and every collapse/split operation, as well as after a group of such modifications.
Similarly, but in reverse, a vertex split 102 takes a single vertex and breaks it into two separate vertices, and an edge and two faces is added between them. In both operations, mesh connectivity is updated to reflect the changes to the mesh, and as discussed below, the effects can be spread over time (geomorphed) to reduce “popping” and “snapping” of mesh alterations.
Preferably, identifying and updating meshes is performed dynamically as viewing parameters change, and is referred herein as view-dependent level-of-detail (LOD) control. The general issue is to locally adjust the complexity of the approximating mesh to satisfy a screen-space pixel tolerance while maintaining a rendered surface that is both spatially and temporally continuous. To be spatially continuous, the mesh should be free of cracks and T-junctions. To be temporally continuous, the rendered mesh should not visibly “pop” from one frame to the next (real-time geomorphs can be used to smooth mesh rendering).
The illustrated edge collapse 100 and vertex split 102 operations involve two vertices vu, vt. These operations are applicable to arbitrary triangle meshes, not just height fields, terrains, or elevation data. However, if one is rendering terrains, special optimizations can be made, such as the omission of normals and other elements in the data structures storing the mesh. (See
In order to simplify a mesh, a set of all possible edge collapses have to be generated and evaluated. For any potential collapse, a heuristic h evaluates the effect of edge collapse operation, and this information is used to identify which collapses result in a mesh “most like” a more detailed mesh.
Comparisons can be made between any mesh LOD, not just an immediately preceding/following mesh level. For example, the heuristic h can compare any potential mesh (defined by proposed edge collapse operations) against the original fully detailed mesh Mn. (Comparing to Mn is the preferred method for approximation-error calculation.)
For terrain meshes, the conventional method of evaluating an edge collapse operation is to apply a heuristic which evaluates approximation error based on maximum vertical deviation at the original grid points, which can result in significant errors. For view dependent LOD of a terrain mesh, such inexact evaluations are unacceptable.
Conventionally, the operation is evaluated at original vertex points 154 and 156 (the first mesh's corresponding vertices in the second mesh). It appears there is an approximation error of zero, because there appears to have been no height deviation resulting from the collapse operation (all vertices have maintained the same height).
But, this is an erroneous result. An inspection of resultant mesh's piecewise interpolant shows a severe alteration. (An interpolant is a point within a plane defined as passing through all vertices of a face.) New interpolated vertices Iw, and Ix originally had height values of zero (mesh 150) and now have respective height values of 1 and 2 (mesh 152). This corresponds to a mesh having substantially different geometry from the original mesh. Even though both interpolants are arguably equally valid, it is clear that the two meshes will render differently. Such errors, even in a single rendering operation, can produce significant pops and visual discontinuities since the height value (as illustrated) can suddenly increase (pop), yet appear to be “optimal” at original vertices.
Thus, a heuristic h which evaluates a collapse by inspecting original vertex points can lead to severely inaccurate rendering. For view dependent LOD of a terrain mesh, this is not tolerable. (See De Floriani et al., Multiresolution models for topographic surface description, The Visual Computer, vol. 12, no. 7, pp. 317-345 (1996); Garland et al., Fast polygonal approximation of terrains and height fields, CMU-CS 95-181, CS Dept., Carnegie Mellon University (1995).)
A better method is to compare all interpolated points to the original fully detailed (Mn) mesh data.
The maximum height deviation between two triangle meshes is known to lie at a vertex of their union partition 162 (not all markers indicated with lead lines) in the plane. An efficient way to enumerate the vertices of the union partition is to consider the grid points adjacent to Vs that are internal to the faces (labeled with squares), and internal to the edges (labeled with circles). The squares identify vertices of the original grid, and the circles identify the intersection points between edges 160 of the reference surface, and edges 164 of a mesh after a proposed edge collapse. The central vertex Vs, by definition, has no error and is labeled with a diamond. (All such marked points are collectively item 162.)
For any two meshes, maximum error has to occur at one of the marked points 162. Therefore, error is computed for each point and the maximum error is simply the largest error value for those points. Error is computed by evaluating the vertical deviation between the reference surface and current mesh; this computed error is not just an upper-bound, it is exact with respect to the original fully detailed mesh. All potential edge collapses are evaluated and the collapse providing the smallest maximum error is selected. This error is stored in a Vsplit field (see
If the original mesh is being partitioned into smaller sub-regions (FIG. 14), a maximum error value may be tracked for all collapse operations within a given region, allowing processing of some blocks to be skipped if the error value indicates an insignificant impact on rendering to an output device.
The complex triangular mesh of
The particular sequence of edge collapse operations must be chosen carefully since they determine the quality of the intermediate levels of detail. Each of the n LOD operations can be stored in data structures (see
Each of the LOD refinements are stacked to form a tree-like vertex hierarchy. Root nodes correspond to the vertices of the base mesh M0, and leaf nodes correspond to the fully detailed mesh Mn. It is important to note that the sequence of vsplit refinements required to perform a view-dependent refinement uniquely define a vertex hierarchy, and permits the creation of selectively refined meshes, or meshes not in the original pre-computed refinement sequence. Although there is a particular ordering to the edge split or collapse operations, dependent on how the coarse mesh M0 was developed, preferably all operations for a given LOD need not be performed: Instead, a vertex front can be defined through the vertex hierarchy. This front defines an active mesh, representing a particular series of edge collapse and vertex split operations.
In a view-dependent progressive-mesh (VDPM) framework, a sequence of refinement transformations uniquely defines a vertex hierarchy 200, in which the root nodes 202-206 correspond to the vertices of a base (simplest) mesh M0, and the leaf nodes 208-224 correspond to the fully detailed mesh. This hierarchy permits the creation of selectively refined meshes, which as discussed above, are meshes not necessarily in the original sequence.
A selectively refined mesh is defined by a “vertex front” 226 across the vertex hierarchy, and is obtained by incrementally applying transformations subject to a set of legality conditions/viewing constraints. Applying a particular sequence of refinements or coarsening operations results in a particular “active mesh,” which is much simpler than the fully detailed mesh Mn. This simplicity allows real-time manipulation of the simpler mesh.
Prior to rendering frame, the active vertex front is traversed, and each vertex is either coarsened or refined based on view-dependent refinement. Refinement is performed if a vertex's neighborhood satisfies predetermined criteria set according to, the type of data being rendered. For large terrain meshes (or large triangle meshes), the preferred criteria is that the neighborhood satisfy 3 requirements: (1) it intersects the view frustum (the region of a mesh visible to a viewer/user), (2) its Gauss map is not strictly oriented away, and (3) its screen-projected deviation from the original mesh exceeds a user-specified pixel tolerance. For efficient and conservative run-time evaluation of these criteria, each vertex in the hierarchy stores the following: a bounding-sphere radius, a normal vector, a cone-of-normals angle, and a deviation space encoded by a uniform component and a directional component. It is understood that other criteria, such as color or other characteristics, may also be stored therein.
It is assumed that the size of an active mesh Mi (having m vertices) is much smaller than that of the fully refined mesh Mn (having n vertices). Therefore, a fully detailed mesh has order n vertices and 2n faces, and the simpler mesh has order m vertices and 2m faces.
A limitation of prior art progressive mesh methods, such as the VDPM taught in U.S. Pat. No. 5,963,209, is that all data structures scale proportionally with the size n of the fully refined mesh. In particular, static storage is allocated to represent the mesh connectivity for all faces in the mesh even though only a small fraction are usually active at any one time. This requires extensive resources in order to compute geomorph transforms. To overcome this problem, instead of allocating resources based on the size of the fully detailed mesh, most allocations are deferred until vertices are used (i.e. become “active”) in processing the mesh.
As shown the structures are separated into two parts: a static part 250 encoding a vertex hierarchy 200 (
The first of the static structures is VGeom 254, which contains the geometry for a point. Elements of the structure are a point 254a of type Point and a normal 254b of type Vector. The point represents the location within 3D space (which can be real-world or unit space), and the normal for the point. VGeom is never allocated on its own, but is used as a type declaration (like ListNode) and is allocated within the Vsplit, AVertex, and VertexMorph structures.
Structure Vertex 256, (of which there are 2n in array vertices 264a), contains pointers 256a, 256b to a dynamically allocated AVertex 266, and its parent Vertex structure, and an index i 256c of the vspliti operation that creates its children. Index i is set to −1 if the vertex is a leaf node of the hierarchy. Since vertices are numbered consecutively, the index i is sufficient to compute the indices of the two child vertices vt and vu and vsplit operation, and of the one or two child faces fl and fr (see FIG. 3). This is accomplished by allocating two Face 258 structures to each vsplit even in the rare case that the vsplit creates only one face.
Structure Face 258, (of which there are 2n in array faces 264b), contains a pointer 258a to a dynamically allocated AFace 268 structure The pointer 258a is set to 0 (or other value indicating non-use) if the face is not active. An active face is one that is presently in the active mesh.
Structure Vsplit 260 (of which there are n in array vsplits 264c) contains a geometry element 260a for a child vertex vu. An array 260b for storing the four neighboring faces fn0, fn1, fn2, fn3 (See
When encoding a terrain mesh, one can optimize the Vgeom and Vsplit structures by discarding the normal 254b, sin2alpha 260d, and uni_error 260e structure elements. (Values that can be discarded when processing terrain meshes are flagged with a † symbol.) The reason is that these normals are usually used in backface simplification and shading operations prior to rendering the mesh. But, by virtue of the viewing perspective on a terrain mesh, most faces are visible and backface simplification is too costly with respect to the number of faces actually removed (i.e., rendering savings are less than computation time).
Normals are not required to map the mesh's texture onto the mesh because a vertical projection is employed. Preferably an image map is used that is the same size as the mesh and therefore texture coordinates can be determined implicitly. When mapping the texture to the mesh, the shape of the mesh is used to warp the texture to conform to the mesh.
Structure Listnode 262 defines a general doubly-linked list. This structure contains pointers to the previous 262a and next 262b nodes in the node list. The list is used both by the Aface and by the AVertex nodes, as well as the head of lists active_vertices 264d and active_faces 264e to keep track of active faces and vertices.
Structure SRMesh 264 corresponds to a particular selectively refinable mesh Mi. Within SRMesh are three arrays and two ListNode lists. The first array 264a, of size order 2n, is an array containing all vertices in the hierarchy. The second array 264b, of size order 2n, is an array containing all faces in Mn. (Note that any selectively refined mesh contains a subset of the faces of Mn) The third array 264c, of size order n, is an array containing the chain of vertex split operations encoding how to generate the completely detailed mesh Mn from current mesh Mi. Recall that the vertex split lists are generated by recording, in reverse order, the generation of the base (i.e., most simplified) mesh M0. The first ListNode list 264d tracks the head of the active vertex list, and the second ListNode list 264e tracks the head of the active face list. Since only 2,000-20,000 triangles may be active out of 17 million triangles in a mesh, only allocating space for active vertices and faces greatly reduces memory consumption.
The dynamically allocated structures 252 require order m storage space (recall that m is the number of active vertices), which is significantly less space then order n (m<n). One can also optimize access as well as storage requirements through register-variable optimizations, optimizing memory boundaries, in-line coding, loop-reductions, etc. For clarity, these compile-time optimizations have been omitted.
The first dynamic structure of
Structure AFace 268, of which there are approximately 2m, tracks the active faces in the selectively refined mesh. This structure contains a ListNode list 268a tracking the list of active faces. An AVertex pointer 268b points to the three vertices, ordered counter-clockwise, that make up the face defined by this structure 268. An AFace pointer 268c points to the (up to) three neighboring faces of the face defined by this structure. Recall that these are the neighboring faces bordering the current triangle face, and they are numbered such that a neighboring face j is located across from (opposite to) a vertex j of the current triangle face.
The final element is an integer 268d referencing a texture tile identifier for the face. Rather than storing textures themselves, only indices 268d into a texture mapping are tracked.
Structure VertexMorph 270, of which there are g (the number of geomorphing vertices), represents each vertex that is in the process of geomorphing. This structure contains a boolean value 270a indicating whether the vertex 256 associated with this structure (Vertex→AVertex→VertexMorph) is undergoing coarsening.
Also within structure VertexMorph 270 is a gtime counter 270b indicating the number of remaining frames in the associated vertex's 256, and two VGeom elements. The first VGeom entry vg_refined 270c stores a backup copy of the morphing vertex's refined geometry, and is used as a temporary placeholder in functions such as vsplit to maintain the vertex's current state before it is altered. One skilled in the art will recognize that the vg_refined 270c variable can be removed from the vsplit function by increasing the complexity of the code.
To enable geomorphs, each active vertex has a field vmorph 266d, which points to a dynamically allocated VertexMorph record 270 when the vertex is morphing. In practice, the number g of morphing vertices is only a fraction of the number m of active vertices, which is itself only a small fraction of the total number n of vertices. (m<n).
Overall, then, the data structures need 88n+112m+52g bytes (i.e., 12(2n)+4(2n)+56n+40m+36(2m)+52g). These requirements are much less than the storage requirements of other methods, which require upwards of 224n memory resources (see, for example, U.S. Pat. No. 5,963,209). In addition, because in practice the number of active faces 2m is generally about 12,000, and 12,000 is less than 65,536, the AVertex* 256a and AFace* 258a pointers in the static structure can be replaced by 16-bit indices. Additionally, in Vsplit 260 we can quantize the coordinates to 16 bits and use 256-entry lookup tables for {rv, uv2, ov2, sin2 αv}.
Static storage is then reduced from 88n to 56n bytes. And, if encoding a terrain mesh, where Vgeom and Vsplit have been optimized by discarding the normal, sin2alpha, and uni_error elements, static storage is reduced to 48n bytes using 16-bit indices, quantization, and lookup tables. In comparison, a standard representation for a pre-simplified, quantized, irregular mesh uses 42n bytes of memory, (n)(12) bytes for positions and normals, and (2n)(3)(4) bytes for connectivity). Thus the illustrated view-dependent progressive-mesh framework only requires a 33% increase in memory over static non-LOD representations.
As compared, for example, with U.S. Pat. No. 5,963,209, geometry storage is considerably reduced by modifying the vsplit/ecol transformations to force vertices vs and vt to have the same geometry, as shown in FIG. 3. This optimization, while optional, results in an average increase of 15% in active faces. Additionally, instead of storing the texture identifiers for the new faces fl and fr, in the Vsplit structure 260 (FIG. 7), they are inferred during a vsplit from the adjacent active faces fn1 and fn3 respectively. (See Vsplit 260 discussion above.)
Shown is pseudocode for applying to the selectively refined mesh the vertex split vspliti that splits a vertex vs into two vertices, vt and vu. (See also
For the new left and right faces fl, fr, at step 304 (the pointers to faces fl and fr are derived from the index vs.i of the vertex split that splits vs, much as above) fl is assigned its value from vs, (indexed into the faces 264b array by vs's index 256c value). Due to how the storage hierarchy is represented in the
So, at step 310 vt acquires vs's avertex pointer reference, but at step 310 vt replaces vs's avertex.vertex back-reference with vt's back-reference. At this point, vs's useful data is exhausted and the vs node is released. Alternatively, rather than creating a new vertex vt and deleting old-vertex vs, the contents of vs can just be modified with vt's new data. For new vertex vu, since vu is not inheriting data from a pre-existing vertex, at step 312 vu is initialized as a new node and added to the vertex list-node tracking list 312. Finally, at step 314 initialization for the left and right faces is completed.
In a highly visual environment, such as a flight simulator, it is crucial that the rendered scenes be presented smoothly to a viewer/user, but without mesh popping. At first, these two goals seem contradictory, since the common method for obtaining high frame rates is (1) dropping intermediate frames, resulting in large “steps” from frame to frame, or (2) maintaining screen-space error tolerance at a value of 1 pixel, with a constant error tolerance, resulting in a greatly varying number of active faces depending on the model complexity near the viewpoint, and a non-uniform frame rate.
It is preferable that a constant frame-rate be maintained. This can be accomplished by adjusting the screen-space error tolerance, and eliminating resulting mesh popping by smoothly morphing the geometry (hence the term “geomorphing”) over time. Therefore, even though the model may (at times) have a projected geometric error of a few pixels, the error is imperceptibly smoothed over a geomorph transition. In addition, because geomorphs hide discontinuities in a mesh, pixel error tolerance can be increased to allow higher frame rates (e.g., 72+frames/sec).
As discussed above, view-dependent LOD results in mesh complexity controlled by predetermined refinement criteria which indicates the need for an edge collapse (ecol) or vertex split (vsplit). To reduce unnecessary popping of a mesh rendering, the rate of mesh transformation (e.g. the number of steps, or length of time to perform the transformation) depends on whether a given portion of an active mesh is entering 360, leaving 362, or staying within 364 the current viewpoint's viewing frustum 354. For mesh portions staying within the frustum 354, instead of performing instantaneous transformations, they are performed as a geomorph by gradually changing the vertex geometry over several frames. A transformation is not performed as a geomorph unless the region of the affected surface is facing the viewer. Note that for height fields (i.e. terrains or elevation data), facial orientation is not determined since vertex normals are discarded to reduce memory requirements. However, this does not impose a significant penalty since surface meshes are viewed essentially horizontally (e.g. one usually looks across the breadth of the mesh) there is only a small fraction of triangles that are completely facing away and unnecessarily rendered. In addition, if face orientation were needed, it can be determined from geometric analysis of vertex positions in the model-space.
A transformation is also not performed unless the affected region (e.g., a vertex neighborhood) overlaps the view frustum 354. It is undesirable to initiate a geomorph on a region 362 known to be invisible, because according to the refinement criteria, such a region may have unbounded screen-space error. (Both the decision to geomorph and the decision to refine the mesh either use or ignore face orientation.) If such a region were to become visible prior to the end of the geomorph, it could lead to an arbitrarily large screen-space displacement. For example, as the frustum 354 pans left, the nearby off-screen region 366 should not be morphing from its coarse state as it enters the left edge 368 of the viewport.
Intersection of a vertex's neighborhood with a viewing frustum can be quickly determined if a bounding-sphere (
And, although the above discussion focuses on manipulating vertex positions, it is understood that besides position information, vertices contain other attributes such as a normal, color, texture coordinates, alpha, blend data, etc. Preferably, normals are interpolated over the unit sphere, and other attributes linearly interpolated. Also, texture coordinates are generated implicitly rather than explicitly during rendering using a linear map on vertex positions. Each AFace structure 268 (
Assuming is_invisible( ) returns false, then a modified vsplit( ) procedure is called to split a node vs. (If “not is_invisible(vs)” evaluates True, then is_Invisible(vs) must evaluate False, i.e., vs is visible.) After execution of the initialization code 300-316, vertex vs is tested to verify it is presently visible 424. If so, the coordinates for vu is assigned the values for vt. If vs is not visible (with respect to a viewing perspective and viewing frustum), then the transformation is not performed as a geomorph over time, and instead the vertex is instantly updated. As with the
A new VertexMorph structure 270 (
The vg_refined 270c backup copy of the refined geometry is set 434 to the child vertex geometry for vs.i, where i is the vsplit index entry for vertex vs. Vg_refined 270c represents the ultimate refined position for a geomorphing vertex, and is used to determine an interpolation increment 436 based on the distance between the starting 438 and ending 270c positions of the morphing vertex. This total distance is then divided by gtime 270b to determine how far the vertex is to move in a given frame.
A key feature of the illustrated geomorph refinement is that the refinement operations may overlap; that is, since vertex connectivity is updated before the geometry is interpolated, a new geomorph process can be applied to a vertex vs already undergoing morphing. Causing a new refinement generates two new nodes vt and vu, as discussed herein, where one node continues along vs's original path, and the other continues on towards the new refinement position for that vertex. When an overlap occurs, the vsplit procedure simply advances the vertex front down the vertex hierarchy (possibly several “layers”), and modifies mesh connectivity instantaneously while deferring geometric changes. Since the refinements can overlap in time, they can be performed in parallel to achieve significant performance gains.
Geomorph coarsening operations reverse a series of refinement operations. As discussed above, when performing a refinement operation, the connectivity of the old and new nodes is updated first, and then the nodes gravitate to their new locations over gtime number of incremental steps. When performing the coarsening operation, similar steps can be implemented to, in effect, reverse a refinement operation.
In the basic case, geomorph coarsening requires that geometry interpolation take place first, and then mesh connectivity can be updated to reflect the changed geometry. This contrasts refinement operations, in which connectivity is updated first (the node is split), and then geometry positions interpolated over gtime. In this configuration, there cannot be multiple overlapping coarsening operations because the connectivity is not updated until after the coarsening operation is complete, resulting in an ancestor-vertices problem. That is, it is impossible to detect whether the second coarsening operation is legal, to determine if the proper neighborhood of faces is present around vs (vt for moment, since vt will become vs) and vw to allow that second edge collapse to occur in the future. (Note that coarsening difficulties are not inherent to the VDPM framework, and also arise in multi-resolution hierarchies based on uniform subdivision.) To overcome this restriction on overlapping geomorph coarsening, as discussed above for
For the illustrated embodiments, a geomorph refinement (apply_vsplit(vs)) is initiated only when the screen-projected deviation of its mesh neighborhood exceeds a pixel tolerance τ (See Hoppe, supra, View-Dependent Refinement of Progressive Meshes, at 193.) This prevents wasting processing resources on rendering faces having too little effect on output. A problem arises in that the viewer is moving, and the mesh neighborhood is likely to be closer to the viewer by the time the geomorph completes; this then invalidates the error estimate. This issue can be resolved by anticipating the viewer location gtime frames into the future, and evaluating the screen-space error metric from that future configuration. This future location can be estimated by extrapolation based on the current per-frame viewer velocity. (A more rigorous, and computation intensive, solution is to account for changes in velocity, and altering the lifetimes of ongoing geomorphs as necessary.)
If these conditions 472 are not met, then a split is not appropriate and checks are made 474 to determine whether an edge collapse is appropriate. If vertex vs has a parent and it is legal to perform an edge collapse, then boolean variable vmc is assigned 476 the logical-and result between active vertex v's current morphing state and whether v is already undergoing a coarsening operation (i.e. vmc is true if and only if the vertex v is currently undergoing geomorph coarsening.) The v's parent vertex v.parent is then checked 478 for visibility. If the parent is invisible (i.e., rendering rejected based upon selected viewing criteria), then if vmc is true, v is allowed to immediately finish its geomorph coarsening transformation 480 (i.e., its position jumps to its final coarsened state). Otherwise, the parent is visible, its screen error is checked 482 to determine if it is greater than error tolerance r. If the parent's error tolerance is more than τ, then there is no point to coarsening the mesh, and the geomorph - coarsening is cancelled 484, thus instantly returning the vertex v to its refined state. Alternatively, if the parent's error tolerance is less than or equal to τ, vmc is true, and if v's morphing gtime 270b (
If none of these conditions apply, then the parent is visible 478, screen error is acceptable 482, and vmc is not true 480 (i.e., no coarsening is being performed).
Therefore, a coarsening geomorph is started 492. Note, however, that this implementation of adapt_refinement( ) does not allow overlapping coarsening geomorphs. (But the
One method for determining screen error is to project a mesh onto a virtual screen space. This projection necessarily converts mesh data from world space coordinates into screen space coordinates. For a regionalized mesh (see
That is, a dynamically structured mesh as disclosed herein can only help so much in the conservation of computer resources. Even simplified meshes may be too large to fit in main memory (item 30, FIG. 1). If one resorts to using a virtual memory manager, the resulting paging operations may cause the process to pause intermittently, disturbing frame rate. By partitioning the refinement database (i.e., the vsplit operations) into a block hierarchy, domain knowledge can be exploited to allow explicit pre-fetching of refinement data before it is needed.
Further, just as the geometry data may be too large for memory, so may its associated texture image. Clip-maps offer an elegant solution to this problem, but require hardware assistance currently available only on high-end systems. A simpler approach is to partition the texture image into tiles that can be mip-mapped and paged independently. In illustrated embodiments, a texture tile may be associated with each block, as the progressive mesh is constructed, so that mesh faces never cross block boundaries.
To perform partitioning, a large hierarchical PM surface geometry 550 is partitioned into many smaller blocks 552, 554, and bottom-up recursion is used to simplify and merge block geometries. The blocks 552, 554 can be arbitrarily shaped (e.g., an Escher design can be used if desired); however, for clarity herein, it is assumed the regions are rectangular. The term “region” is used to represent the arbitrarily shaped partitions of the mesh. In defining the regions, no mesh faces are allowed to cross partition boundaries. This restriction allows the regions to be independently processed, since all ancestor/descendent vertices will always be contained within a given region's ancestral tree within the PM hierarchy.
The advantage to breaking a mesh into sub-regions is that that memory requirements can be arbitrarily reduced depending on the initial region (e.g., block) size. For computers providing a limited resource for storing the mesh, small regions can be defined, which as discussed below, are recursively coarsened and combined until a final simple mesh is obtained.
In addition, as discussed above regarding the creation of the progressive mesh, the maximum error for a given region of the mesh may be tracked and associated with the region. Then, during rendering/display of the progressive mesh, the region's maximum error can be compared against a particular screen error tolerance (see
If the Stop State test 500 fails, then we have not reached the “bottom” of the recursion process, and we need to further divide the input mesh. The current input mesh is broken into multiple smaller regions M1, M2, M3, and M4, and the Partition Simplify( ) function is recursively called 506, 508 to evaluate each of M1-M4. Note that in a multi-processing/multitasking environment, each partition request can be assigned to a different execution thread.
However, if the Stop State test 500 succeeds, then we are at the bottom of the recursion process, which means that the input region has reached a predetermined minimal size, or has satisfied some other stop state condition. The input mesh is then simplified 502, and the simplification steps recorded for storage in the progressive mesh hierarchy. Now we begin to fall back out of the recursion process, so that the rest of the mesh can be simplified down to the base mesh. (Note that although this description assumes a recursive partition function, it is understood that equivalent iterative functions may be used instead.)
After falling back from the function call 508 to partition M4, the next step is to recombine 510 the now-simplified versions of M1-M4 back into the originally larger region. (If a multi-processing or multi-threading implementation is used, one may need to block after processing M4 and wait until M1-M3 are completed.) After re-combining, this larger mesh region is then simplified and the simplification operations recorded for storage in the refinement list for the progressive mesh hierarchy.
As the system falls out of the recursion process, each row of a progressive mesh hierarchy will be generated according to the recursion pattern. When the original recursive calls are reached, the original four mesh regions are combined and simplified to form M0, the base mesh. Since all simplifications were recorded as the system fell out of the recursion process, a fully defined progressive mesh is achieved with arbitrarily minimal memory requirements (e.g. if the stop state is a very small mesh size, then few resources will be required to process it). Note that it is assumed in the pseudo-code that the structures storing the input mesh M are replaced with the combination of M1-M4. One can also design the functions so that only pointers to the portions of a mesh are passed during sub-division, so that all simplifications are performed in-place.
Continuing now with
For each simplified block 556, 558, a sequence of transformations 560, 562 are applied to the blocks, resulting in simpler blocks 564, 566. (Note that the arrow connecting blocks 556, 558 to respective blocks 564, 566 represents the coarsening transformations.) As discussed for
Final blocks are then stitched together into a combined block 574. Note that the
During the simplification process, refinement dependencies between adjacent blocks can be avoided by constraining transformations to leave boundary vertices untouched. Similarly, if texture tiles are desired, transformations can be constrained to prevent displacement of tile boundaries within a block. And as with the first level, special treatment is given to the last level of simplification. In the last level, in which there is only a single block 580, simplification of the mesh boundary is permitted since inter-block dependencies are now immaterial. Thus, the final base mesh consists of only 2 triangles, or 2 triangles per tile if texture tiles are used.
Except for the first row 592 of the hierarchy containing base mesh M0, each successive row corresponds to those transformations that are required to obtain an ith level of detail of the original fully detailed mesh Mn. Each row is organized according to the blocks defined in the original mesh, hence row contents vsplits A 596 corresponds to refining block 564 (
Visually, each successive row groups refinement operations for more detailed blocks below and within the bounds for the simplified combination of the detailed blocks (e.g., split operations 602, 604 are below and within vsplit A 596) So, the second row contains vsplits S 594, which contains the final transformation required to transform the last stitched-together block 574 (
As mentioned above, a user specified threshold determines the maximum approximation error at each simplification level. These thresholds can be chosen so that block refinements at all levels have roughly the same number of refinements. Since the thresholds form an upper-bound on the errors of all transformation below that level, they can be used to determine which block refinements need to be memory-resident based on the current viewing parameters, and which others should be pre-fetched based on the anticipated view changes. Specifically, we can use the thresholds to compute the maximum screen-projected error for each active block. Thus, if the error exceeds the screen-space tolerance, its child block refinements are loaded into memory and further tested.
And, since block refinements correspond to contiguous sequences in the vertices, faces, and vsplits data structures (FIG. 7), we can reserve virtual memory for the whole PM (i.e., the entire arrays), and use sparse memory allocation to commit only a fraction to physical memory. For example, the Microsoft Windows operating system supports such a reserve/commit/decommit protocol using the VirtualAlloc( ) and VirtualFree( ) system calls. (See generally Microsoft Win32 Programmer's Reference, vol. 2, chap. 42, Microsoft Press (1993).)
The invention teaches avoiding inefficiencies inherent to prior art progressive mesh rendering systems. In particular, discussion has focused on relieving computation requirements by breaking down complex meshes into multiple smaller regions, thereby reducing a potentially intractable problem into manageable portions. Having described and illustrated the principles of our invention with reference to an illustrated embodiment, it will be recognized that the illustrated embodiment can be modified in arrangement and detail without departing from such principles. The programs, processes, or methods described herein are not related or limited to any particular type of computer apparatus, unless indicated otherwise. Various types of general purpose or specialized computer apparatus may be used with or perform operations in accordance with the teachings described herein. Elements of the illustrated embodiment shown in software may be implemented in hardware and vice versa, and it should be recognized that the detailed embodiments are illustrative only and should not be taken as limiting the scope of our invention. What is claimed as the invention are all such embodiments as may come within the scope and spirit of the following claims and equivalents thereto.
Patent | Priority | Assignee | Title |
10043309, | Dec 14 2015 | Microsoft Technology Licensing, LLC | Maintaining consistent boundaries in parallel mesh simplification |
10096150, | Jun 17 2014 | Imagination Technologies Limited | Assigning primitives to tiles in a graphics processing system |
10313673, | Oct 19 2016 | GOOGLE LLC | Methods and apparatus to encode and/or decode normals of geometric representations of surfaces |
10430975, | Nov 17 2016 | GOOGLE LLC | Advanced k-D tree encoding for point clouds by most significant axis selection |
10496336, | Nov 17 2016 | GOOGLE LLC | K-D tree encoding for point clouds using deviations |
10553035, | Jun 02 2017 | GOOGLE LLC | Valence based implicit traversal for improved compression of triangular meshes |
10692275, | Jun 17 2014 | Imagination Technologies Limited | Assigning primitives to tiles in a graphics processing system |
10733766, | Oct 19 2016 | GOOGLE LLC | Methods and apparatus to encode and/or decode normals of geometric representations of surfaces |
10891758, | Jul 23 2018 | GOOGLE LLC | Geometry encoder |
10950042, | Jun 02 2017 | GOOGLE LLC | Guided traversal in compression of triangular meshes |
10957097, | May 29 2014 | Imagination Technologies Limited | Allocation of primitives to primitive blocks |
10977773, | Feb 19 2019 | GOOGLE LLC | Cost-driven framework for progressive compression of textured meshes |
11017580, | Jun 28 2018 | BEIJING BYTEDANCE NETWORK TECHNOLOGY CO , LTD | Face image processing based on key point detection |
11195325, | Jul 01 2015 | 3D4Medical Limited | Method and apparatus for freeform cutting of digital three dimensional structures |
11217016, | Nov 24 2020 | Electronic Arts Inc.; ELECTRONIC ARTS INC | Systems and methods for generating a proxy mesh for an artist-authored mesh using conformed smoothing |
11244498, | Jun 17 2014 | Imagination Technologies Limited | Assigning primitives to tiles in a graphics processing system |
11481952, | May 29 2014 | Imagination Technologies Limited | Allocation of primitives to primitive blocks |
11527041, | Jan 24 2019 | Canon Kabushiki Kaisha | Information processing apparatus, information processing method and storage medium |
11568575, | Feb 19 2019 | GOOGLE LLC | Cost-driven framework for progressive compression of textured meshes |
11763524, | Sep 13 2019 | ANSYS, Inc | Layered meshing for additive manufacturing simulations |
11889072, | Dec 28 2018 | Hangzhou Hikvision Digital Technology Co., Ltd. | Video encoding and decoding |
6995761, | Jan 14 2000 | California Institute of Technology | Compression of 3D surfaces using progressive geometry |
7002574, | Dec 27 2002 | Microsoft Technology Licensing, LLC | Method and system for tessellating a polygon |
7038700, | Sep 26 2001 | Mazda Motor Corporation | Morphing method for structure shape, its computer program, and computer-readable storage medium |
7221372, | Sep 13 2002 | ENTROPIC COMMUNICATIONS, INC ; Entropic Communications, LLC | Method of analyzing and modifying a footprint |
7299165, | Oct 16 2003 | SIEMENS PRODUCT LIFECYCLE MANAGEMENT SOFTWARE INC | System, method, and computer program product for identifying manufacturing-specific regions in object models |
7372466, | Mar 12 2002 | Sony Corporation | Image processing apparatus and method of same |
7426292, | Aug 07 2003 | Mitsubishi Electric Research Laboratories, Inc. | Method for determining optimal viewpoints for 3D face modeling and face recognition |
7499599, | Jun 12 2003 | General Electric Company | Method of real-time correction of non-functioning pixels in digital radiography |
7539606, | Nov 15 2001 | NINTENDO CO , LTD | System and method for efficiently simulating and imaging realistic water surface and other effects |
7663625, | Mar 23 2001 | Dassault Systemes | Collaborative design |
7680350, | May 07 2004 | TERRAMETRICS, LLC | Method and system for progressive mesh storage and reconstruction using wavelet-encoded height fields |
7733338, | Jul 20 2004 | AUTODESK, Inc | Reduction of a mesh with preservation of flow lines |
7899241, | May 07 2004 | TERRAMETRICS, LLC | Method and system for progressive mesh storage and reconstruction using wavelet-encoded height fields |
7924278, | Jul 28 2006 | Microsoft Technology Licensing, LLC | Real-time GPU rendering of piecewise algebraic surfaces |
8120607, | May 30 2008 | Nvidia Corporation | Boundary transition region stitching for tessellation |
8373717, | Apr 25 2007 | Nvidia Corporation | Utilization of symmetrical properties in rendering |
8411105, | May 14 2004 | Nvidia Corporation | Method and system for computing pixel parameters |
8416242, | May 14 2004 | Nvidia Corporation | Method and system for interpolating level-of-detail in graphics processors |
8432394, | May 14 2004 | Nvidia Corporation | Method and system for implementing clamped z value interpolation in a raster stage of a graphics pipeline |
8441497, | Aug 07 2007 | Nvidia Corporation | Interpolation of vertex attributes in a graphics processor |
8725476, | May 04 2010 | Lucasfilm Entertainment Company Ltd. | Applying details in a simulation |
8749576, | May 14 2004 | Nvidia Corporation | Method and system for implementing multiple high precision and low precision interpolators for a graphics pipeline |
8825452, | Mar 05 2010 | Omron Corporation | Model producing apparatus, model producing method, and computer-readable recording medium in which model producing program is stored |
8970592, | Apr 19 2011 | Lucasfilm Entertainment Company LLC | Simulating an arbitrary number of particles |
8972230, | Nov 15 2001 | NINTENDO CO , LTD | System and method for using indirect texturing to efficiently simulate and image surface coatings and other effects |
9123160, | Oct 30 2011 | Lockheed Martin Corporation | Concurrent mesh generation in a computer simulation |
9123183, | Oct 30 2011 | Lockheed Martin Corporation | Multi-layer digital elevation model |
9147283, | Oct 30 2011 | Lockhead Martin Corporation; Lockheed Martin Corporation | Water surface visualization during a simulation |
9161012, | Nov 17 2011 | ZHIGU HOLDINGS LIMITED | Video compression using virtual skeleton |
9165403, | Jan 14 2011 | Intel Corporation | Planetary scale object rendering |
9342918, | Nov 15 2001 | Nintendo Co., Ltd. | System and method for using indirect texturing to efficiently simulate and image surface coatings and other effects |
9773341, | Mar 14 2013 | Nvidia Corporation | Rendering cover geometry without internal edges |
9787321, | Nov 17 2016 | GOOGLE LLC | Point cloud data compression using a space-filling curve |
9965893, | Jun 25 2013 | GOOGLE LLC | Curvature-driven normal interpolation for shading applications |
RE42406, | Jun 18 1998 | HANGER SOLUTIONS, LLC | Multi-resolution geometry |
Patent | Priority | Assignee | Title |
5883629, | Jun 28 1996 | International Business Machines Corporation | Recursive and anisotropic method and article of manufacture for generating a balanced computer representation of an object |
6009435, | Nov 21 1997 | GOOGLE LLC | Progressive compression of clustered multi-resolution polygonal models |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Jul 14 1998 | Microsoft Corporation | (assignment on the face of the patent) | / | |||
Jul 14 1998 | HOPPE, HUGUES H | Microsoft Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 009505 | /0885 | |
Oct 14 2014 | Microsoft Corporation | Microsoft Technology Licensing, LLC | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 034541 | /0001 |
Date | Maintenance Fee Events |
Sep 22 2008 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Oct 28 2008 | ASPN: Payor Number Assigned. |
Sep 27 2012 | M1552: Payment of Maintenance Fee, 8th Year, Large Entity. |
Sep 29 2016 | M1553: Payment of Maintenance Fee, 12th Year, Large Entity. |
Date | Maintenance Schedule |
Apr 12 2008 | 4 years fee payment window open |
Oct 12 2008 | 6 months grace period start (w surcharge) |
Apr 12 2009 | patent expiry (for year 4) |
Apr 12 2011 | 2 years to revive unintentionally abandoned end. (for year 4) |
Apr 12 2012 | 8 years fee payment window open |
Oct 12 2012 | 6 months grace period start (w surcharge) |
Apr 12 2013 | patent expiry (for year 8) |
Apr 12 2015 | 2 years to revive unintentionally abandoned end. (for year 8) |
Apr 12 2016 | 12 years fee payment window open |
Oct 12 2016 | 6 months grace period start (w surcharge) |
Apr 12 2017 | patent expiry (for year 12) |
Apr 12 2019 | 2 years to revive unintentionally abandoned end. (for year 12) |